feat: enhance calibration figure with Brier score and update prediction functions

app/training/generate_figures.py

@@ -239,26 +239,35 @@ def fig_feature_importance(results: dict, top_n: int = 20) -> None:
     print("  ✓ feature_importance_top20.png")
 
 
-def fig_calibration(results: dict) -> None:
+def fig_calibration(results: dict, y_true: np.ndarray, y_prob: np.ndarray) -> None:
     """Calibration curve — does predicted probability match reality?"""
     fig, ax = plt.subplots(figsize=(7, 6.5))
     best = results.get("_best_model", "XGBoost")
-    items = [(k, v) for k, v in results.items() if not k.startswith("_") and isinstance(v, dict)]
-
-    colors = plt.cm.plasma(np.linspace(0.2, 0.8, len(items)))
-    for idx, (name, data) in enumerate(items):
-        y_true = np.array(data["y_true"])
-        y_prob = np.array(data["y_prob"])
-        frac_pos, mean_pred = calibration_curve(y_true, y_prob, n_bins=10)
-        lw = 3 if name == best else 1.2
-        ax.plot(mean_pred, frac_pos, "o-", color=colors[idx], lw=lw,
-                label=f"{name}", markersize=6 if name == best else 4)
+    frac_pos, mean_pred = calibration_curve(y_true, y_prob, n_bins=10)
 
+    ax.plot(mean_pred, frac_pos, "o-", color=PALETTE["primary"], lw=3,
+            markersize=8, label=f"{best}")
+    ax.fill_between(mean_pred, frac_pos, mean_pred, alpha=0.15,
+                    color=PALETTE["primary"])
     ax.plot([0, 1], [0, 1], "k:", alpha=0.5, label="Mükemmel / Perfect")
+
+    # Brier score annotation
+    brier = float(np.mean((y_prob - y_true) ** 2))
+    ax.text(
+        0.04, 0.94,
+        f"Brier Score = {brier:.4f}\nN = {len(y_true)} (5-fold CV)",
+        transform=ax.transAxes,
+        fontsize=10, va="top",
+        bbox=dict(boxstyle="round,pad=0.5", facecolor=PALETTE["bg"],
+                  edgecolor=PALETTE["primary"], alpha=0.85),
+    )
+
     ax.set_xlabel("Ortalama Tahmin Olasılığı / Mean Predicted Probability")
     ax.set_ylabel("Gerçek Pozitif Oranı / Fraction of Positives")
-    ax.set_title("Kalibrasyon Eğrisi", fontsize=13, fontweight="bold")
-    ax.legend(loc="upper left", framealpha=0.85, fontsize=9)
+    ax.set_title("Kalibrasyon Eğrisi — En İyi Model", fontsize=13, fontweight="bold")
+    ax.legend(loc="lower right", framealpha=0.85, fontsize=10)
+    ax.set_xlim([0, 1])
+    ax.set_ylim([0, 1])
     plt.savefig(FIGURES_DIR / "calibration_plot.png")
     plt.close()
     print("  ✓ calibration_plot.png")
@@ -393,17 +402,24 @@ def main() -> None:
         importance.items(), key=lambda x: x[1], reverse=True,
     )]
 
+    print("\nLoading dataset...")
+    X, y = _load_csv_data(feature_cols)
+    X_scaled = scaler.transform(X)
+
+    print("Computing 5-fold cross-validated predictions (this may take ~1-2 min)...")
+    cache: dict = {}
+    y_true, y_pred, y_prob = _get_cv_predictions(model, X_scaled, y, cache)
+
     print("\nGenerating figures...")
-    fig_confusion_matrix(results)
-    fig_roc_comparison(results)
-    fig_pr_curves(results)
+    fig_confusion_matrix(results, y_true, y_pred)
+    fig_roc_comparison(results, y_true, y_prob)
+    fig_pr_curves(results, y_true, y_prob)
     fig_feature_importance(results)
-    fig_calibration(results)
+    fig_calibration(results, y_true, y_prob)
     fig_model_comparison(results)
     fig_feature_distributions(feature_cols, top_features)
 
-    print("\nLoading data for SHAP (this may take ~30s)...")
-    X, y = _load_csv_data(feature_cols)
+    print("\nGenerating SHAP summary...")
     fig_shap_summary(model, scaler, feature_cols, X)
 
     print(f"\nDone. {len(list(FIGURES_DIR.glob('*.png')))} figures in {FIGURES_DIR}")